21.02.08 · number-theory / quadratic-forms-local-fields

Hasse-Minkowski theorem

shipped3 tiersLean: none

Anchor (Master): Minkowski 1890 *J. reine angew. Math.* 106 (originator, rational forms); Hasse 1923 *J. reine angew. Math.* 152 + Hasse 1924 *J. reine angew. Math.* 153 (representation local-global, modern statement); Serre 1973 *A Course in Arithmetic* Ch. III-IV (the canonical short modern proof using the Hilbert symbol product formula); Cassels 1978 *Rational Quadratic Forms* Ch. 6 (full textbook proof); O'Meara 1963 *Introduction to Quadratic Forms* (Springer Grundlehren 117) Ch. VI (general global-field setting); Lam 2005 *Introduction to Quadratic Forms over Fields* (AMS GSM 67) Ch. VI; Meyer 1884 *J. reine angew. Math.* 98 (indefinite forms in five or more variables); Selmer 1951 *Acta Mathematica* 85 (cubic counterexample); Manin 1971 ICM Nice (Brauer-Manin obstruction)

Intuition Beginner

A Diophantine question asks whether a polynomial equation in several variables has a rational solution. The Hasse-Minkowski theorem says that for quadratic equations the question splits cleanly into smaller pieces. You check whether the equation has a real solution; you check whether it has a $2$ -adic solution, a $3$ -adic solution, a $5$ -adic solution, and so on for every prime. If the answer is yes at every place, then the original equation has a rational solution. If the answer is no at any single place, then it has none.

Why is this remarkable? The completions $Q_{p}$ are simpler than $Q$ itself. Solving an equation over $Q_{p}$ reduces to solving a sequence of congruences modulo $p, p^{2}, p^{3}, \dots$ , with a built-in mechanism (Hensel's lemma) that lifts approximate solutions to exact ones. Over $R$ the test is just a sign computation. So a question about rational solutions becomes a finite list of local checks, each of which is computable.

This is the foundational example of what is called the local-global principle, and quadratic forms are the case where it works perfectly. For cubic equations the principle can fail, and the gap between local solvability and global solvability becomes one of the central subjects of modern arithmetic geometry. The theorem builds toward Brauer-Manin obstructions, the Hasse-Weil $L$ -function, and the Langlands programme, all of which trace their origin to this clean local-global statement for quadratics.

Visual Beginner

A schematic showing the rational number field $Q$ at the centre, with arrows fanning outward to its completions: $R = Q_{\infty}$ at one corner, and the $p$ -adic fields $Q_{2}, Q_{3}, Q_{5}, Q_{7}, \dots$ at the others. Each arrow is labelled "local check". A single line at the top of the picture states the theorem: a quadratic form represents zero over $Q$ if and only if it represents zero over every completion.

The picture captures the essential structure. A single global question (does the form have a rational zero?) becomes an infinite family of local questions, but only finitely many of them are non-redundant — for almost every prime $p$ the answer is automatic from the coefficients of the form. The theorem is the bridge that allows one to assemble local information into a global conclusion.

Worked example Beginner

Check whether the quadratic form $f (x, y, z) = 3 x^{2} + 5 y^{2} - 7 z^{2}$ has a non-zero rational solution to $f = 0$ , by working through each local condition.

Step 1. Check the real place. Over $R$ , the form has signature $(2, 1)$ : two positive coefficients, one negative. An indefinite real quadratic form in three variables always represents zero — pick $x = y = 0$ and any non-zero $z$ and the form is $- 7 z^{2}$ , which can be made any negative value, while choosing $z = 0$ and varying $x, y$ produces any non-negative value, so by continuity zero is reached. So the real condition holds.

Step 2. Check the prime $p = 3$ . We ask whether the equation $3 x^{2} + 5 y^{2} - 7 z^{2} = 0$ has a non-zero solution in $Q_{3}$ . Modulo $3$ the form is $0 \cdot x^{2} + 2 y^{2} - z^{2} \equiv 2 y^{2} - z^{2} (mod 3)$ . A solution with $y \neq = 0$ or $z \neq = 0$ modulo $3$ would force $z^{2} \equiv 2 y^{2} (mod 3)$ , so $(z / y)^{2} \equiv 2 (mod 3)$ . The squares modulo $3$ are ${0, 1}$ and $2$ is not among them. So a solution must have $y \equiv z \equiv 0 (mod 3)$ , which forces $x \neq = 0 (mod 3)$ .

Substituting $y = 3 y^{'}$ , $z = 3 z^{'}$ into the original equation gives $3 x^{2} + 45 y^{'2} - 63 z^{'2} = 0$ , or $x^{2} + 15 y^{'2} - 21 z^{'2} = 0$ after dividing by $3$ . Modulo $3$ this is $x^{2} \equiv 0$ , forcing $x \equiv 0 (mod 3)$ . That contradicts $x \neq = 0 (mod 3)$ from the previous step, so no $3$ -adic solution exists.

Step 3. Conclude. The $3$ -adic condition fails. By Hasse-Minkowski, the form does not represent zero over $Q$ either. So there is no non-zero rational $(x, y, z)$ with $3 x^{2} + 5 y^{2} - 7 z^{2} = 0$ .

Step 4. Confirm by direct search. A small-coefficient search over $Z$ with $∣ x ∣, ∣ y ∣, ∣ z ∣ \leq 10$ finds no integer solutions; a denominator clear from any rational solution would produce an integer solution, so the absence at small heights is consistent with the local obstruction.

What this tells us: a single local obstruction at one prime is enough to rule out rational solvability. The theorem reduces a search problem (try every rational triple) to a finite list of decidable local conditions. The example also shows why the theorem is computationally powerful — checking the $3$ -adic condition took one page of arithmetic.

Check your understanding Beginner

Exercise (easy, numeric).

For how many places $v$ of $Q$ do we need to check the local condition in the Hasse-Minkowski theorem?

Hint

The places of $Q$ are the real place $v = \infty$ plus one for each prime. For a fixed form, almost all $p$ -adic conditions are automatic, but in principle the theorem asks about every place.

Answer

Infinitely many in principle. The places of $Q$ are ${\infty, 2, 3, 5, 7, \dots}$ , one for the real completion and one for each rational prime. In practice, for a fixed quadratic form, almost every $p$ -adic check is automatic: the form has good reduction at every prime not dividing its discriminant, and Hensel's lemma forces representability there. So one checks the real place plus the primes dividing the discriminant — a finite list. Rubric: full credit for "infinitely many in principle, finitely many in practice".

Formal definition Intermediate+

The setting is a non-degenerate quadratic form $f \in Q [x_{1}, \dots, x_{n}]$ , by which we mean a homogeneous polynomial of degree two that, after diagonalisation over $Q$ , has all coefficients non-zero: $f \sim a_{1} x_{1}^{2} + \dots + a_{n} x_{n}^{2}$ with $a_{i} \in Q^{\times}$ . The form represents zero by a non-zero vector over a field $K \supseteq Q$ if there exist $(x_{1}, \dots, x_{n}) \in K^{n} ∖ {(0, \dots, 0)}$ with $f (x_{1}, \dots, x_{n}) = 0$ .

The places of $Q$ are the equivalence classes of non-zero absolute values, classified by Ostrowski: one archimedean place $v = \infty$ giving the usual absolute value $∣ \cdot ∣_{\infty}$ on $Q$ , and one non-archimedean place for each prime $p$ giving the $p$ -adic absolute value $∣ p^{k} \cdot u / v ∣_{p} = p^{- k}$ for $u, v$ coprime to $p$ . The completion of $Q$ at $v$ is $Q_{v}$ : $Q_{\infty} = R$ for the archimedean place, $Q_{p}$ the $p$ -adic field for the non-archimedean ones.

The Hilbert symbol $(a, b)_{v} \in {\pm 1}$ at a place $v$ is defined for $a, b \in Q_{v}^{\times}$ by $$ (a, b)_v = \begin{cases} +1 & \text{if } z^2 = a x^2 + b y^2 \text{ has a non-zero solution in } \mathbb{Q}_v^3, \ -1 & \text{otherwise.} \end{cases} $$ The symbol is bilinear and symmetric, depends only on the classes of $a, b$ in $Q_{v}^{\times} / (Q_{v}^{\times})^{2}$ , and is non-degenerate: for every non-square $a \in Q_{v}^{\times}$ there is a $b$ with $(a, b)_{v} = - 1$ .

The product formula (Hilbert reciprocity): for every $a, b \in Q^{\times}$ , $$ \prod_v (a, b)_v = 1, $$ with the product over all places of $Q$ . Only finitely many factors are not $+ 1$ (those at the primes dividing $ab$ and at $v = \infty$ ), so the product is well-defined. The product formula is itself an elementary version of quadratic reciprocity, and is the bridge between the local Hilbert symbols and the global structure of $Q$ .

The Hasse-Witt invariant $ε_{v} (f) \in {\pm 1}$ of a diagonalised form $f = a_{1} x_{1}^{2} + \dots + a_{n} x_{n}^{2}$ at a place $v$ is $$ \varepsilon_v(f) = \prod_{i < j} (a_i, a_j)_v. $$ It is invariant under equivalence of $f$ over $Q_{v}$ , and together with the rank, the discriminant $d (f) = a_{1} \dots a_{n} \in Q_{v}^{\times} / (Q_{v}^{\times})^{2}$ , and (for $v = \infty$ ) the signature, classifies $f$ up to equivalence over $Q_{v}$ (Serre Ch. IV Theorem 7).

Counterexamples to common slips

Degeneracy of $f$ is excluded. A degenerate form like $f = x^{2} - x^{2} = 0$ represents zero by construction, and the local-global question is vacuous. The theorem is stated for non-degenerate forms; degenerate forms are split off as a hyperbolic plane $H = x^{2} - y^{2}$ plus a non-degenerate complement.
The theorem is about quadratic forms specifically. For cubic forms and higher degrees, local-global can fail: Selmer's $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ has rational points over every $Q_{v}$ but no non-zero rational point. The cubic counterexample is canonical and dates to 1951.
The local condition is required at every place, not just the real place. The form $3 x^{2} + 5 y^{2} - 7 z^{2}$ has a real zero but no $3$ -adic zero, hence no rational zero. Forgetting any single place breaks the implication.

Key theorem with proof Intermediate+

Theorem (Hasse-Minkowski; Serre Ch. IV Theorem 8). Let $f \in Q [x_{1}, \dots, x_{n}]$ be a non-degenerate quadratic form. The form $f$ represents zero by a non-zero vector over $Q$ if and only if it represents zero by a non-zero vector over every completion $Q_{v}$ , where $v$ ranges over the places ${\infty, 2, 3, 5, 7, \dots}$ of $Q$ .

Proof. The forward direction is immediate: a non-zero rational solution is also a non-zero $Q_{v}$ -solution at every place. The substance is the converse. We argue case-by-case on $n$ .

Case $n = 2$ . Diagonalise as $f = a x^{2} + b y^{2}$ with $a, b \in Q^{\times}$ . The form represents zero by a non-zero vector over a field $K$ iff $- b / a \in K^{\times 2}$ . So the local hypothesis says $- b / a$ is a square in every $Q_{v}$ , and we must show $- b / a$ is a square in $Q$ . By unique factorisation in $Z$ , every $r \in Q^{\times}$ has a canonical form $r = \pm \prod_{p} p^{e_{p} (r)}$ with $e_{p} (r) \in Z$ . The element $r$ is a square in $Q_{p}$ iff $e_{p} (r)$ is even and (for $p = 2$ ) the unit part of $r$ is a square in $Z_{2}^{\times}$ ; over $R$ the element $r$ is a square iff $r \geq 0$ . If $r$ is a square in every $Q_{v}$ , then every $e_{p} (r)$ is even and $r > 0$ over $R$ , so $r = \prod_{p} p^{e_{p} (r)}$ with $e_{p} (r) \in 2 Z$ , and $r \in Q^{\times}$ . This completes $n = 2$ .

Case $n = 3$ . After diagonalisation and rescaling, we may write $f \sim x^{2} - a y^{2} - b z^{2}$ with $a, b \in Q^{\times}$ squarefree. The form represents zero by a non-zero vector over a field $K$ iff there exist $(x, y, z) \in K^{3} ∖ {0}$ with $x^{2} = a y^{2} + b z^{2}$ , which is to say the Hilbert symbol $(a, b)_{K} = + 1$ in the sense of the formal-definition section (with $K$ in place of $Q_{v}$ ). The local hypothesis is $(a, b)_{v} = + 1$ for every place $v$ of $Q$ . The conclusion to prove is that $f$ represents zero over $Q$ , equivalently $(a, b)_{Q} = + 1$ in the appropriate global sense, equivalently there is a non-zero rational $(x, y, z)$ with $x^{2} - a y^{2} - b z^{2} = 0$ .

The key analytic input is Dirichlet's theorem on primes in arithmetic progressions: for every modulus $m \geq 1$ and residue class $r$ coprime to $m$ , there are infinitely many primes $p \equiv r (mod m)$ . We use Dirichlet to produce auxiliary primes in carefully chosen congruence classes.

Reduce to the case where $a, b$ are squarefree integers, by multiplying out denominators (which only changes $f$ by squares) and absorbing square factors. Let $S$ be the finite set of places consisting of $\infty$ , $2$ , and every prime dividing $ab$ . At each $v \in S$ the hypothesis $(a, b)_{v} = + 1$ gives a local solution to $x^{2} - a y^{2} - b z^{2} = 0$ in $Q_{v}$ .

The argument now proceeds by induction on $∣ a ∣ + ∣ b ∣$ . Without loss of generality $∣ a ∣ \leq ∣ b ∣$ . If $∣ b ∣ = 1$ then $b = \pm 1$ . The case $b = 1$ gives $x^{2} - a y^{2} - z^{2} = 0$ , immediately solvable by $(x, y, z) = (1, 0, 1)$ . The case $b = - 1$ gives $x^{2} - a y^{2} + z^{2} = 0$ , which requires $- a$ to be a sum of two rational squares; this follows from the local conditions at $\infty$ (forcing $a > 0$ for indefiniteness) and at every $p$ via the two-squares theorem and Hilbert reciprocity. If $∣ a ∣ = ∣ b ∣ = 1$ then $f \sim x^{2} \pm y^{2} \pm z^{2}$ , easily handled directly.

For the inductive step, write $b = \pm p_{1} \dots p_{k}$ as a product of distinct primes. Pick one prime $p_{i}$ dividing $b$ . We claim $a$ is a square modulo $p_{i}$ . The argument: at $v = p_{i}$ the Hilbert symbol $(a, b)_{p_{i}} = + 1$ means $x^{2} - a y^{2} - b z^{2} = 0$ has a non-zero $p_{i}$ -adic solution. Since $p_{i} ∣ b$ , modulo $p_{i}$ this becomes $x^{2} \equiv a y^{2} (mod p_{i})$ , with $(x, y) \neq \equiv (0, 0) (mod p_{i})$ (otherwise the leading $p_{i}$ -adic valuations of $x$ and $y$ both rise, contradicting the choice of a primitive solution). So $a \equiv (x / y)^{2} (mod p_{i})$ is a square modulo $p_{i}$ .

By the Chinese remainder theorem, choose an integer $t$ such that $t \equiv a (mod p_{i})$ for each $p_{i} ∣ b$ and $∣ t ∣ < ∣ b ∣/2$ . Then $t^{2} \equiv a (mod b)$ , so $t^{2} - a = b \cdot b^{'}$ for some integer $b^{'}$ with $∣ b^{'} ∣ < ∣ b ∣$ . Now $f$ represents zero over a field $K$ iff $g := x^{2} - a y^{2} - b^{'} z^{2}$ does, since the substitution $z = z^{'} / t$ in the relation $t^{2} - a = b b^{'}$ exchanges the two representability questions. By the inductive hypothesis applied to $g$ (with $∣ b^{'} ∣ < ∣ b ∣$ , so $∣ a ∣ + ∣ b^{'} ∣ < ∣ a ∣ + ∣ b ∣$ ), the form $g$ represents zero over $Q$ , and hence so does $f$ .

The use of Dirichlet's theorem comes in the refinement of the inductive step when one needs to choose an auxiliary prime $ℓ$ in a prescribed congruence class to make the Hilbert symbols at the modified set $S^{'}$ behave correctly. Specifically, the product formula $\prod_{v} (a, b)_{v} = 1$ forces the parity of the bad places to be even, and Dirichlet supplies the prime needed to balance the count. This is the load-bearing reduction step in Serre Ch. IV §3.

Case $n = 4$ . Diagonalise $f \sim a_{1} x_{1}^{2} + a_{2} x_{2}^{2} + a_{3} x_{3}^{2} + a_{4} x_{4}^{2}$ . Split as $f = g + h$ with $g = a_{1} x_{1}^{2} + a_{2} x_{2}^{2}$ and $h = - (- a_{3} x_{3}^{2} - a_{4} x_{4}^{2})$ . The form $f$ represents zero iff $g$ and $- h = a_{3} x_{3}^{2} + a_{4} x_{4}^{2}$ represent a common value $c \in Q$ . The local hypothesis at each place $v$ gives a common value $c_{v} \in Q_{v}$ . We must show there is a single $c \in Q$ representable by both $g$ and $- h$ over $Q$ .

The set of values represented by a binary form over $Q_{v}$ is described by Hilbert-symbol conditions: $c \in Q_{v}^{\times}$ is represented by $a x^{2} + b y^{2}$ iff $(c, - ab)_{v} = (a, b)_{v}$ , a relation in the group $Q_{v}^{\times} / (Q_{v}^{\times})^{2}$ . So the common-value question at each place is a system of Hilbert-symbol equations, and the case $n = 3$ supplies a global $c$ that satisfies the system at every $v \in S$ simultaneously. With $c$ chosen, the form $g - c$ represents zero in three variables, as does $- h - c$ separately; by the $n = 3$ case applied to each, the global representability follows. This is the counting argument Serre presents on Ch. IV §3 pp. 41-42; it is the only place where the proof becomes genuinely combinatorial rather than algebraic.

Case $n \geq 5$ . Induction reduces the $n = 5$ case to the $n = 4$ case after splitting off a binary form, and the same reduction continues for $n \geq 6$ . The base case for the induction is Meyer's theorem: every indefinite quadratic form in $\geq 5$ variables over $Q$ represents zero by a non-zero vector. The definite case in $\geq 5$ variables over $Q$ has no real zeros (the form is positive- or negative-definite, all coefficients of the same sign), so the local hypothesis already fails. Hence the only forms to consider in $n \geq 5$ are indefinite, and Meyer's theorem closes the induction. $□$

Bridge. Hasse-Minkowski builds toward every later local-global statement in arithmetic geometry, and the foundational reason it holds is exactly the Hilbert symbol product formula $\prod_{v} (a, b)_{v} = 1$ together with the local classification of quadratic forms by rank, discriminant, signature, and Hasse-Witt invariant. The central insight is that the global Hilbert symbol $(a, b)_{Q}$ is determined by its local components, and the product formula forces an even number of local symbols to be $- 1$ ; the local-global hypothesis pins every component to $+ 1$ , hence the global symbol is $+ 1$ and the form represents zero. This is exactly the calculation that appears again in 21.05.01 $ℓ$ -adic Galois representations, where local Tate cohomology and the global Brauer-Hasse-Noether theorem replay the same reciprocity at higher rank. The bridge is the recognition that putting these together identifies the Hasse principle for quadratic forms with the vanishing of a certain cohomology class in $H^{2} (Q, μ_{2})$ , and this cohomological framing generalises to the modern Hasse principle for Brauer-Severi varieties and torsors under semisimple algebraic groups. The local-global identifies $Q$ -solvability with $A_{Q}$ -solvability for quadratics, and the failure of this identification for cubics is exactly what the Brauer-Manin obstruction measures.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Compute the Hilbert symbol $(2, 3)_{5}$ at the prime $5$ , and explain the calculation.

Hint

The symbol $(a, b)_{p}$ for $p$ odd, $p ∤ ab$ , is computed by the Legendre-symbol formula $(a, b)_{p} = (\frac{a}{p})^{v_{p} (b)} (\frac{b}{p})^{v_{p} (a)}$ (Serre Ch. III, Theorem 1). Here $5 ∤ 6$ , so both Legendre exponents are zero and the symbol is $+ 1$ .

Answer

$(2, 3)_{5} = + 1$ . Since $5 ∤ 2$ and $5 ∤ 3$ , the $5$ -adic valuations $v_{5} (2) = v_{5} (3) = 0$ . By Serre Ch. III Theorem 1, for odd $p$ with $p ∤ ab$ the symbol $(a, b)_{p}$ equals $+ 1$ identically (the tame formula has both Legendre-exponents zero).

Concretely, the equation $x^{2} - 2 y^{2} - 3 z^{2} = 0$ has a $5$ -adic solution. Set $y = z = 1$ : $2 + 3 = 5 \equiv 0 (mod 5)$ , so the reduction is $x^{2} \equiv 0 (mod 5)$ with $x = 0$ as the modulo- $5$ root. Hensel's lemma lifts this to an exact $Z_{5}$ -solution because the partial derivative $\partial_{y} (x^{2} - 2 y^{2} - 3 z^{2}) = - 4 y$ is a $5$ -adic unit at $y = 1$ . So $(0, 1, 1)$ is a $5$ -adic point and the symbol value is $+ 1$ . Rubric: full credit for $+ 1$ with any correct justification.

Exercise 4 (medium, symbolic).

Use Hasse-Minkowski to decide whether $f (x, y, z) = x^{2} + y^{2} - 7 z^{2}$ represents zero over $Q$ .

Hint

Check the real place (signature $(2, 1)$ , indefinite, so yes). Then check each prime dividing the discriminant. For the prime $7$ , reduce modulo $7$ and ask whether $x^{2} + y^{2} \equiv 0$ has a non-zero solution.

Answer

No, the form does not represent zero. Real place: signature $(2, 1)$ , indefinite, so represents zero over $R$ — that condition holds. Prime $7$ : modulo $7$ the form reduces to $x^{2} + y^{2} \equiv 0 (mod 7)$ , asking whether $- 1$ is a square in $F_{7}$ . The Legendre symbol $(\frac{- 1}{7}) = (- 1)^{(7 - 1) /2} = (- 1)^{3} = - 1$ , so $- 1$ is not a square modulo $7$ , and the only modulo- $7$ solution is $x \equiv y \equiv 0$ . Substituting $x = 7 x^{'}$ , $y = 7 y^{'}$ gives $49 x^{'2} + 49 y^{'2} - 7 z^{2} = 0$ , equivalently $7 x^{'2} + 7 y^{'2} - z^{2} = 0$ ; modulo $7$ this forces $z \equiv 0$ , breaking primitivity. So no $Q_{7}$ -solution exists, and by Hasse-Minkowski no rational zero exists either. Rubric: full credit for "no" with the modulo- $7$ Legendre-symbol computation.

Exercise 5 (medium, short-answer).

State and prove the local-global theorem for squares: a rational number $r \in Q^{\times}$ is a square in $Q$ iff it is a square in every $Q_{v}$ .

Hint

This is the $n = 2$ case of Hasse-Minkowski applied to the form $r x^{2} - y^{2}$ . Use unique factorisation in $Z$ to track parities of prime exponents.

Answer

Statement. For $r \in Q^{\times}$ , $r \in (Q^{\times})^{2}$ iff $r \in (Q_{v}^{\times})^{2}$ for every place $v$ of $Q$ . Proof. The forward direction is immediate. For the converse, write $r = \pm \prod_{p} p^{e_{p}}$ with $e_{p} \in Z$ uniquely determined by unique factorisation. Over $R$ , $r$ a square forces $r > 0$ , so the sign is $+$ . At each prime $p$ , $r$ a square in $Q_{p}$ forces $v_{p} (r) = e_{p}$ to be even (since the $p$ -adic valuation of a square is twice an integer). For $p = 2$ there is the additional unit condition: the unit part of $r$ at $2$ must be a square in $Z_{2}^{\times}$ , which by the structure of $Z_{2}^{\times} = {\pm 1} \times (1 + 4 Z_{2})$ is the condition that the leading $2$ -adic unit is congruent to $1$ modulo $8$ . With all $e_{p}$ even and the sign positive, $r = \prod_{p} p^{e_{p}}$ is the square of $\prod_{p} p^{e_{p} /2} \in Q^{\times}$ . $□$ . Rubric: full credit for the unique-factorisation argument tracking each prime exponent.

Exercise 6 (medium, symbolic).

State the product formula for the Hilbert symbol and verify it for $a = 3$ , $b = 5$ at $Q$ .

Hint

The product formula is $\prod_{v} (a, b)_{v} = 1$ over all places. Compute each $(3, 5)_{v}$ for $v \in {\infty, 2, 3, 5}$ and check the product is $+ 1$ .

Answer

Product formula: $\prod_{v} (a, b)_{v} = 1$ for every $a, b \in Q^{\times}$ . Verification for $(a, b) = (3, 5)$ . At $v = \infty$ : both $a, b > 0$ , so $(3, 5)_{\infty} = + 1$ . At $v = 2$ : the symbol $(3, 5)_{2} = (- 1)^{((3 - 1) /2) ((5 - 1) /2)} = (- 1)^{1 \cdot 2} = + 1$ by the Serre Ch. III tame formula at $p = 2$ for two odd units. At $v = 3$ : $v_{3} (3) = 1$ , $v_{3} (5) = 0$ , so the Serre Ch. III formula gives $(3, 5)_{3} = (\frac{5}{3}) = (\frac{2}{3}) = - 1$ . At $v = 5$ : $v_{5} (3) = 0$ , $v_{5} (5) = 1$ , so $(3, 5)_{5} = (\frac{3}{5}) = - 1$ (since $3$ is not a square in $F_{5}$ , where the squares are ${1, 4}$ ). At any other prime $p$ neither divides $3$ nor $5$ , so $(3, 5)_{p} = + 1$ . Product: $(+ 1) (+ 1) (- 1) (- 1) = + 1$ . $□$ . Rubric: full credit for the four-place computation and the product equalling $+ 1$ .

Exercise 7 (hard, short-answer).

State Meyer's theorem and explain how it closes the $n \geq 5$ case of Hasse-Minkowski.

Hint

Meyer's theorem (1884): an indefinite quadratic form in $\geq 5$ variables over $Q$ represents zero by a non-zero vector. The Hasse-Minkowski induction at $n \geq 5$ splits the form into a binary plus an $(n - 2)$ -variable piece and applies the $n = 4$ result.

Answer

Meyer's theorem (1884): every indefinite non-degenerate quadratic form in $\geq 5$ variables with rational coefficients represents zero by a non-zero vector over $Q$ . Closing the $n \geq 5$ case of Hasse-Minkowski. Given a form $f$ in $n \geq 5$ variables satisfying the local hypothesis at every place, we want to conclude $f$ represents zero over $Q$ . If $f$ is positive- or negative-definite over $R$ , the real-place hypothesis $f$ represents zero in $R^{n}$ already fails — the only real zero of a positive-definite form is the origin. So $f$ is indefinite over $R$ . Meyer's theorem then directly gives a non-zero rational zero, with no recourse to the $p$ -adic conditions at all. So the case $n \geq 5$ has a stronger conclusion: the real local hypothesis alone (plus indefiniteness) suffices when $n \geq 5$ . The original proof of Meyer's theorem uses geometry of numbers; modern proofs derive it from the $n = 4$ case of Hasse-Minkowski by splitting off a hyperbolic plane and reducing the variable count. Rubric: full credit for stating Meyer and explaining that the indefinite local-real-place hypothesis is what closes the case.

Exercise 8 (hard, short-answer).

Selmer's curve $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ has rational points over every $Q_{v}$ but no non-zero rational point. Explain in two paragraphs why this does not contradict Hasse-Minkowski, and name the cohomological invariant that detects the failure.

Hint

Hasse-Minkowski is a statement specifically about quadratic forms. Selmer's curve is a cubic, so the theorem does not apply. The Brauer-Manin obstruction is the modern tool that captures the cubic failure.

Answer

Why no contradiction. Hasse-Minkowski is a theorem about non-degenerate quadratic forms; its statement and proof both depend essentially on the structure of $Q^{\times} / (Q^{\times})^{2}$ , the Hilbert symbol, and the bilinear / quadratic-form machinery. Selmer's $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ is a smooth cubic curve, and the analogous local-global principle (every Diophantine equation that has local points over every $Q_{v}$ has a rational point) is the Hasse principle, which is much stronger than Hasse-Minkowski and is false in general for cubic and higher-degree equations. Selmer 1951 produced this cubic precisely to show the Hasse principle fails; the rational-point obstruction is real, but it does not contradict any theorem about quadratic forms. The cohomological invariant. The Brauer-Manin obstruction (Manin 1971 ICM) refines the adelic point set $X (A_{Q})$ by a pairing with the Brauer group $Br (X) \to Br (A_{Q})$ , producing the Brauer-Manin set $X (A_{Q})^{Br} \subseteq X (A_{Q})$ . Any rational point lies in this Brauer-Manin set. For Selmer's curve, $X (A_{Q})^{Br} = \emptyset$ while $X (A_{Q}) \neq = \emptyset$ , so the obstruction is non-zero and accounts for the absence of rational points. The Brauer-Manin obstruction is the cohomological invariant that captures the failure of the Hasse principle for many (though not all) higher-genus curves and varieties; the question of whether Brauer-Manin is the only obstruction for curves is an active research area (Skorobogatov 1999 produced examples where it is not). Rubric: full credit for (i) cubic vs quadratic distinction; (ii) naming the Brauer-Manin obstruction.

Advanced results Master

Theorem (Hasse-Minkowski over a number field). Let $K$ be a global field and $f$ a non-degenerate quadratic form over $K$ in $n \geq 1$ variables. Then $f$ represents zero by a non-zero vector over $K$ iff $f$ represents zero by a non-zero vector over every completion $K_{v}$ , where $v$ ranges over the places of $K$ .

The proof for general $K$ follows the same case-by-case structure as the proof over $Q$ . The case $n = 2$ uses the structure of $K^{\times} / (K^{\times})^{2}$ in place of $Q^{\times} / (Q^{\times})^{2}$ . The case $n = 3$ uses the Hilbert reciprocity law $\prod_{v} (a, b)_{v} = 1$ over the places of $K$ , which is a consequence of class field theory (in particular the global reciprocity map and its compatibility with local reciprocity). The cases $n = 4$ and $n \geq 5$ run identically. The number-field generalisation is the standard target for treatments at full generality (O'Meara 1963 Ch. VI).

Theorem (classification of quadratic forms over $Q$ ). Two non-degenerate quadratic forms $f, g \in Q [x_{1}, \dots, x_{n}]$ are equivalent over $Q$ iff they are equivalent over every $Q_{v}$ . Equivalence over $Q_{v}$ is determined by the rank, the discriminant $d (f) \in Q_{v}^{\times} / (Q_{v}^{\times})^{2}$ , the Hasse-Witt invariant $ε_{v} (f) \in {\pm 1}$ , and (for $v = \infty$ ) the signature $(p, q)$ .

This is Minkowski's original 1890 theorem, sharpened by Hasse 1923. The local invariants at each finite place are computable from the diagonalisation, and the only constraint linking them globally is the product formula $\prod_{v} ε_{v} (f) = 1$ and the analogous relation for the discriminant. The Hasse principle for equivalence is the equivalence-of-forms version of the representation theorem above.

Theorem (Brauer-Manin obstruction; Manin 1971 ICM). Let $X$ be a smooth projective variety over $Q$ and $Br (X)$ its (cohomological) Brauer group. There is a natural pairing $$ X(\mathbb{A}_\mathbb{Q}) \times \mathrm{Br}(X) \to \mathbb{Q}/\mathbb{Z}, \qquad ((P_v), A) \mapsto \sum_v \mathrm{inv}v(A(P_v)), $$ *where $inv_{v} : Br (Q_{v}) \to Q / Z$ is the local invariant of class field theory and $A (P_{v}) \in Br (Q_{v})$ is the specialisation. Any rational point $P \in X (Q)$ pairs to zero with every $A \in Br (X)$ ; the Brauer-Manin set $X (A_{Q})^{Br}$ is the set of adelic points that pair to zero with every Brauer element. The inclusion $X (Q) \subseteq X (A_{Q})^{Br}$ is automatic; Selmer's cubic curve has $X (A_{Q})^{Br} = \emptyset$ while $X(\mathbb{A}\mathbb{Q}) \neq \emptyset$, demonstrating the obstruction is non-zero.*

For smooth projective varieties of higher genus, the Brauer-Manin obstruction is conjectured (Colliot-Thélène, Sansuc) to be the only obstruction for the Hasse principle; this is open in general and was sharpened in the period 1990-2010 with the introduction of higher descent and étale-Brauer refinements. For curves of genus $\geq 1$ , Skorobogatov 1999 produced examples where Brauer-Manin is insufficient, opening the study of higher cohomological obstructions.

Theorem (Hasse principle for number fields). For a number field $K$ , the analogue of Hasse-Minkowski holds: a non-degenerate quadratic form over $K$ represents zero by a non-zero vector over $K$ iff it does so over every $K_{v}$ . More generally, for any global field $K$ and any algebraic group $G$ that is semisimple, simply connected, and connected over $K$ , the Hasse principle holds for $G$ -torsors: a $G$ -torsor over $K$ is split iff it is split over every $K_{v}$ .

The general Hasse principle for simply connected semisimple groups is the Kneser-Harder theorem (Kneser 1965 for type $A_{n}$ , Harder 1968 for all classical types, completed by Chernousov 1989 for type $E_{8}$ ). The principle fails outside the simply-connected setting: for the adjoint form of $E_{7}$ over a number field, there are non-split torsors that are locally split everywhere. The Hasse-Minkowski theorem for quadratic forms is the foundational case of this much wider family of local-global statements, and the bridge between them is the recognition that a quadratic form is a torsor under an orthogonal group.

Theorem (Hasse-Minkowski via Witt rings). The Witt ring $W (Q)$ of non-degenerate quadratic forms over $Q$ modulo hyperbolic forms is determined by the family of local Witt rings ${W (Q_{v}) : v}$ : the natural map $W (Q) \to \prod_{v} W (Q_{v})$ is injective, with image determined by the product-formula condition $\prod_{v} ε_{v} = 1$ on the second-cohomology piece.

The Witt-ring formulation is Lam's 2005 modern packaging. The injectivity of $W (Q) \to \prod_{v} W (Q_{v})$ is exactly the equivalence statement of Hasse-Minkowski; the image description is the additional global constraint coming from Hilbert reciprocity. The Witt-ring formulation makes the algebraic content of Hasse-Minkowski transparent: it is the injectivity of a localisation map for a Witt-theoretic functor on global fields, with image cut out by the Hilbert reciprocity relations.

Synthesis. Hasse-Minkowski is the foundational reason every classical local-global statement in arithmetic geometry has the shape it has, and the central insight is that the Hilbert symbol product formula $\prod_{v} (a, b)_{v} = 1$ is exactly the analytic-algebraic mechanism by which local information assembles into global solvability. Putting these together, the Hasse principle for quadratic forms, the Hasse principle for Brauer-Severi varieties, the Kneser-Harder Hasse principle for simply connected semisimple groups, and the Brauer-Manin obstruction for general varieties form one local-global framework whose first non-zero instance is the three-variable case of Hasse-Minkowski itself. This pattern recurs in the Tate-Shafarevich group of an elliptic curve, where locally-everywhere-vanishing cohomology classes form the obstruction to a Hasse principle for principal homogeneous spaces; in the Langlands programme, where local Langlands at each place assembles into a global Langlands correspondence (when the strong Hasse principle conjectures hold); and in the étale cohomology of varieties over global fields, where the Brauer-Manin pairing and its refinements measure the deviation from a local-global principle for rational points.

The bridge between the elementary Hilbert-symbol calculus of Serre Ch. III and the cohomological local-global theory of class field theory is the recognition that the product formula is a degree-two statement about the Brauer group: $Br (Q) \to ⨁_{v} Br (Q_{v}) \to Q / Z \to 0$ is exact, with the second map summing local invariants, and Hasse-Minkowski for ternary forms is the statement that the kernel of the first map is captured by the local invariants modulo this sum-to-zero relation. The bridge is exactly the Albert-Brauer-Hasse-Noether theorem that this exact sequence is the global Brauer-group sequence, identifying the elementary local-global theory of quadratic forms with the cohomological local-global theory of central simple algebras over $Q$ .

Full proof set Master

Proposition 1 (case $n = 2$ of Hasse-Minkowski). A binary form $a x^{2} + b y^{2}$ with $a, b \in Q^{\times}$ represents zero over $Q$ iff $- b / a$ is a square in every $Q_{v}$ .

Proof. The form represents zero iff there exist $(x, y) \neq = (0, 0)$ in $Q^{2}$ with $a x^{2} = - b y^{2}$ , equivalently $- b / a = (x / y)^{2}$ over $Q$ . So representability is equivalent to $- b / a \in (Q^{\times})^{2}$ . By the local-global theorem for squares (Exercise 5 above), $- b / a$ is a square in $Q$ iff it is a square in every $Q_{v}$ . The argument uses unique factorisation in $Z$ : the $p$ -adic valuation of a rational square is twice an integer, and the sign is positive over $R$ ; the conjunction of these conditions at every place forces a global square root via the canonical factorisation $r = \pm \prod_{p} p^{e_{p}}$ with all $e_{p}$ even and the sign positive. $□$

Proposition 2 (case $n = 3$ of Hasse-Minkowski; Serre Ch. IV Theorem 8 ternary case). A ternary form $x^{2} - a y^{2} - b z^{2}$ with $a, b \in Q^{\times}$ represents zero over $Q$ iff the Hilbert symbol $(a, b)_{v} = + 1$ for every place $v$ of $Q$ .

Proof. The forward direction: if $x^{2} - a y^{2} - b z^{2} = 0$ has a non-zero solution over $Q$ , it has a non-zero solution over each $Q_{v}$ , so $(a, b)_{v} = + 1$ at every $v$ by definition of the Hilbert symbol.

The converse is the substantive direction. Assume $(a, b)_{v} = + 1$ for every $v$ . Reduce $a, b$ to squarefree integers as in the theorem proof. Induct on $∣ a ∣ + ∣ b ∣$ . The base cases $∣ a ∣ + ∣ b ∣ \leq 2$ are handled directly. For the inductive step, choose a prime $p ∣ b$ . The local hypothesis $(a, b)_{p} = + 1$ provides a $p$ -adic solution $(x_{0}, y_{0}, z_{0})$ to $x^{2} - a y^{2} - b z^{2} = 0$ , primitive (rescale so that $min (v_{p} (x_{0}), v_{p} (y_{0}), v_{p} (z_{0})) = 0$ ). Reducing modulo $p$ kills the $b z^{2}$ term, leaving $x_{0}^{2} \equiv a y_{0}^{2} (mod p)$ . Primitivity forces $(x_{0}, y_{0}) \neq \equiv (0, 0) (mod p)$ . If $y_{0} \equiv 0 (mod p)$ then $x_{0} \equiv 0$ , contradicting primitivity; so $y_{0} \neq \equiv 0$ , and $a \equiv (x_{0} / y_{0})^{2} (mod p)$ , making $a$ a square modulo $p$ .

By the Chinese remainder theorem, choose an integer $t$ with $t \equiv a (mod p)$ for every prime $p ∣ b$ , and $∣ t ∣ \leq ∣ b ∣/2$ . Then $t^{2} - a \equiv 0 (mod b)$ , so $t^{2} - a = b b^{'}$ for some $b^{'} \in Z$ with $∣ b^{'} ∣ \leq (t^{2} + ∣ a ∣) /∣ b ∣ \leq (b^{2} /4 + ∣ a ∣) /∣ b ∣ < ∣ b ∣$ when $∣ a ∣ \leq ∣ b ∣$ (the strict inequality holds for $∣ b ∣ \geq 3$ ; small cases are checked directly).

The form $f = x^{2} - a y^{2} - b z^{2}$ represents zero over a field $K$ iff $f^{'} = x^{2} - a y^{2} - b^{'} z^{2}$ does. This follows from the chain of equivalences: a non-zero zero of $f$ is a triple $(x, y, z)$ with $x^{2} - a y^{2} = b z^{2}$ , and using $b b^{'} = t^{2} - a$ we can rewrite $b z^{2} (t^{2} - a) = b^{'} (b z)^{2} (t^{2} - a) / b^{'} = (b z)^{2} t^{2} - a (b z)^{2} = (b z t)^{2} - a (b z)^{2}$ . Re-grouping, $f$ represents zero iff $f^{'}$ does, after a change of variables encoded by the Witt-ring identity $⟨ 1, - a, - b ⟩ \sim ⟨ 1, - a, - b^{'} ⟩$ in $W (K)$ whenever $b b^{'} (t^{2} - a) \in K^{\times 2}$ .

By the inductive hypothesis with $∣ a ∣ + ∣ b^{'} ∣ < ∣ a ∣ + ∣ b ∣$ , the form $f^{'}$ represents zero over $Q$ once we verify the local hypothesis $(a, b^{'})_{v} = + 1$ at every place $v$ . The product formula and bilinearity of the Hilbert symbol give $(a, b^{'})_{v} = (a, b)_{v} \cdot (a, b b^{'})_{v}^{- 1} = (a, b)_{v} \cdot (a, t^{2} - a)_{v}^{- 1}$ . Now $(a, t^{2} - a)_{v} = (a, - a)_{v} \cdot (a, 1 - t^{2} / a)_{v}$ , and $(a, - a)_{v} = + 1$ identically (one of the basic Hilbert-symbol identities; Serre Ch. III §1 Proposition 2), and $(a, 1 - t^{2} / a)_{v}$ is also $+ 1$ because $1 - t^{2} / a$ differs from a square by a small perturbation. So $(a, b^{'})_{v} = (a, b)_{v} = + 1$ , and the inductive hypothesis applies. The required Dirichlet-on-primes-in-arithmetic-progressions input enters in the choice of the auxiliary prime needed to balance the parity of ${(a, b^{'})_{v}}$ over the place set $S \cup S^{'}$ , where $S = {\infty} \cup {p : p ∣ 2 ab}$ is the original bad set and $S^{'}$ is the modified set after replacing $b$ by $b^{'}$ . Dirichlet supplies a prime $ℓ \equiv c (mod m)$ for a specific congruence class, and $ℓ$ enters into the modified diagonal to make the global symbols agree. This completes the induction. $□$

Proposition 3 (case $n = 4$ of Hasse-Minkowski). A quaternary form $a_{1} x_{1}^{2} + a_{2} x_{2}^{2} + a_{3} x_{3}^{2} + a_{4} x_{4}^{2}$ with $a_{i} \in Q^{\times}$ represents zero over $Q$ iff it represents zero over every $Q_{v}$ .

Proof. The forward direction is automatic. For the converse, split as $g (x_{1}, x_{2}) + h (x_{3}, x_{4})$ with $g = a_{1} x_{1}^{2} + a_{2} x_{2}^{2}$ and $h = a_{3} x_{3}^{2} + a_{4} x_{4}^{2}$ . The form $f = g + h$ represents zero by a non-zero vector iff there is a $c \in Q$ represented both by $g$ (so $g (x_{1}, x_{2}) = c$ for some $(x_{1}, x_{2}) \neq = 0$ ) and by $- h$ (so $- h (x_{3}, x_{4}) = c$ for some $(x_{3}, x_{4}) \neq = 0$ ), with the convention that "represented by" includes $c = 0$ in which case $g$ and $- h$ each represent zero separately.

At each place $v$ , the local hypothesis gives a non-zero $Q_{v}$ -solution and hence a $c_{v} \in Q_{v}$ represented by both $g$ and $- h$ in $Q_{v}$ . The set of values $c \in Q_{v}^{\times}$ represented by $g$ over $Q_{v}$ is described by the Hilbert-symbol condition $(c, - a_{1} a_{2})_{v} = (a_{1}, a_{2})_{v}$ , an equation in $Q_{v}^{\times} / (Q_{v}^{\times})^{2}$ . Similarly for $- h$ : $(c, - a_{3} a_{4})_{v}^{- 1} = (- 1, - 1)_{v} (a_{3}, a_{4})_{v}$ (using the symbol algebra rules). Together these are two conditions on $c \in Q_{v}^{\times} / (Q_{v}^{\times})^{2}$ , forming a $F_{2}$ -affine subspace at each finite place.

The global existence of $c \in Q^{\times}$ satisfying the system at every place is a Hasse-Minkowski-style local-global question on a ternary form: think of $(c, - a_{1} a_{2})$ and $(c, - a_{3} a_{4})$ as Hilbert pairings, and the joint condition ${(c, - a_{1} a_{2})_{v} = (a_{1}, a_{2})_{v}} \cap {(c, - a_{3} a_{4})_{v} = (a_{3}, a_{4})_{v} (- 1, - 1)_{v}}$ as defining a non-empty subset of $Q^{\times} / (Q^{\times})^{2}$ via the ternary Hasse-Minkowski just proved. With $c$ produced, $g (x_{1}, x_{2}) = c$ and $- h (x_{3}, x_{4}) = c$ each represent zero (in three variables after adjoining the value $c$ ), so by ternary Hasse-Minkowski each has a $Q$ -solution, and the combined $(x_{1}, x_{2}, x_{3}, x_{4})$ is a non-zero rational zero of $f$ .

The technical care needed is the counting / measure argument that the joint condition has non-empty intersection over $Q$ — this requires Dirichlet's theorem in the form that there exist primes representing prescribed Hilbert-symbol patterns, plus the product formula constraining the parity. Serre Ch. IV §3 pp. 41-42 carries this out, with the key step being a counting argument on the number of primes in a fixed Hilbert-symbol class. $□$

Proposition 4 (Meyer's theorem; Meyer 1884). Every indefinite non-degenerate quadratic form $f \in Q [x_{1}, \dots, x_{n}]$ in $n \geq 5$ variables represents zero by a non-zero vector over $Q$ .

Proof sketch. Meyer's original 1884 proof uses geometry of numbers (Minkowski's lattice-point theorem) on the indefinite form. Modern proofs reduce to the $n = 4$ Hasse-Minkowski case as follows. An indefinite form $f$ in $n \geq 5$ variables has signature $(p, q)$ with $p, q \geq 1$ . Split off a hyperbolic plane $x_{1} x_{2}$ from the diagonalisation, leaving an $(n - 2)$ -variable form $f^{'}$ in $n - 2 \geq 3$ variables. The original $f$ represents zero iff $f^{'}$ does (the hyperbolic plane always represents zero). For $n = 5$ the residual form is ternary and Hasse-Minkowski $n = 3$ applies; for $n \geq 6$ iterate. The local conditions at each $Q_{v}$ are satisfied because the indefiniteness over $R$ propagates to a non-zero local zero at every place (using the fact that quadratic forms in $\geq 5$ variables over $Q_{p}$ for $p$ odd always represent zero — this is the elementary fact that a non-degenerate form in $\geq 5$ variables over a $p$ -adic field is universal). The detailed reduction is in Serre Ch. IV §4. $□$

Proposition 5 (Selmer's counterexample; Selmer 1951). The smooth cubic curve $C : 3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ in $P_{Q}^{2}$ has $C (Q_{v}) \neq = \emptyset$ for every place $v$ of $Q$ but $C (Q) = {(0, 0, 0)}$ , i.e. no non-zero rational point.

Proof sketch. Local solvability at every place is checked by Hensel's lemma at each prime (the curve has good reduction at every $p \neq = 2, 3, 5$ , where smooth-mod- $p$ points lift), plus explicit small computations at $p = 2, 3, 5$ and the real place. The absence of non-zero rational points is the substantive content of Selmer 1951; his proof uses class-field-theoretic descent on the elliptic curve to which $C$ is birationally equivalent (an isogeny to the elliptic curve $E : y^{2} = x^{3} - 432 \cdot 15$ in Weierstrass form, with $C$ a $3$ -isogenous twist). The descent computation shows that the $3$ -Selmer group $Sel_{3} (E / Q)$ is non-zero but $E (Q) = {O}$ (rank zero), forcing a non-zero element of the Tate-Shafarevich group $\Sha (E / Q) [3]$ . This element represents $C$ itself, and the local vanishing of the $3$ -Selmer class is the local-everywhere-rational-point condition on $C$ , while non-zeroity in $\Sha$ is exactly the absence of global rational points. The Brauer-Manin obstruction subsequently identifies the same class via the pairing $C (A_{Q}) \times Br (C) \to Q / Z$ . See Cassels 1962 Arithmetic on curves of genus 1, IV for the modern reading of Selmer's argument. $□$

Proposition 6 (Hilbert product formula). For every $a, b \in Q^{\times}$ , $\prod_{v} (a, b)_{v} = 1$ , with the product over all places of $Q$ .

Proof. By bilinearity and symmetry of the Hilbert symbol, reduce to the case $a, b \in {- 1, p}$ for primes $p$ . There are three cases: $(- 1, - 1)$ , $(- 1, p)$ , and $(p, q)$ for primes $p, q$ . Each is evaluated at every place using Serre Ch. III Theorem 1 (the explicit tame and wild formulas), and the product is checked to be $+ 1$ in each case. The case $(p, q)$ for $p, q$ odd distinct primes is equivalent to the quadratic reciprocity law $(\frac{p}{q}) (\frac{q}{p}) = (- 1)^{(p - 1) (q - 1) /4}$ , so the product formula is a packaging of quadratic reciprocity. The cases involving $- 1$ recover the first and second supplementary laws. Serre Ch. III Theorem 3 gives the full case-by-case argument. $□$

Connections Master

Dirichlet $L$ -functions and primes in arithmetic progressions 21.03.02. The proof of Hasse-Minkowski at $n = 3$ uses Dirichlet's theorem on primes in arithmetic progressions in a key reduction step: to balance the parity of Hilbert symbols at a modified place set, the proof needs a prime in a prescribed congruence class. Dirichlet's theorem, proved via the non-vanishing of $L (1, χ)$ for non-principal characters, is the load-bearing analytic input. Without it, the inductive step at $n = 3$ does not close, and the entire Hasse-Minkowski proof depends on the existence of primes in arithmetic progressions.
$ℓ$ -adic Galois representations 21.05.01. The Hilbert symbol and the global product formula are degree-two facts about $Br (Q)$ , and the cohomological formulation $0 \to Br (Q) \to ⨁_{v} Br (Q_{v}) \to Q / Z \to 0$ is the Albert-Brauer-Hasse-Noether sequence. This sequence reads as a local-global statement on the Galois cohomology $H^{2} (Q, G_{m})$ via Tate's theorem, which is exactly the setting of $ℓ$ -adic Galois representations of $Gal (\overline{Q} / Q)$ . Hasse-Minkowski is the prototypical instance of the local-global principle that the Langlands programme generalises to higher-rank automorphic representations.
Riemann zeta function 21.03.01. Dirichlet's theorem on primes in arithmetic progressions, the analytic input to the $n = 3$ case, is a consequence of the non-vanishing of Dirichlet $L$ -functions at $s = 1$ , which in turn rests on the analytic continuation and functional equation of $ζ (s)$ . The chain Hasse-Minkowski $\leftarrow$ Dirichlet $\leftarrow$ $ζ$ is one of the cleanest examples of the way analytic and algebraic number theory interlock: the existence of primes in congruence classes is an analytic question (about $L$ -series at $s = 1$ ), and its solution feeds into a purely algebraic local-global theorem about quadratic forms.
Bilinear and quadratic forms 01.01.15. The objects whose local-global theory is governed by Hasse-Minkowski. The foundational unit develops the diagonalisation, discriminant, signature, and Sylvester's law of inertia; Hasse-Minkowski is the arithmetic refinement that adds the Hasse-Witt invariant $ε_{v}$ at each finite place and the local-global classification statement. The connection runs both ways: every theorem about rational quadratic forms invokes Hasse-Minkowski to translate global statements into a finite list of local invariants.
Modularity and BSD 21.06.01. The Brauer-Manin obstruction (Manin 1971), introduced to explain the failure of the Hasse principle for higher-degree forms like Selmer's cubic, is closely tied to the Tate-Shafarevich group of an elliptic curve and to the BSD conjecture. For an elliptic curve $E / Q$ , the Tate-Shafarevich group $\Sha (E / Q)$ measures the failure of the Hasse principle for principal homogeneous spaces under $E$ , and its conjectured finiteness (a piece of refined BSD) is the natural higher-rank refinement of the quadratic-form local-global theorem. Hasse-Minkowski sits at the foundational level of this hierarchy.

Historical & philosophical context Master

Legendre's 1785 Mémoires sur l'analyse indéterminée ^{[Legendre 1785]} proved the ternary case of Hasse-Minkowski in the disguise of Legendre's three-square theorem: the equation $a x^{2} + b y^{2} + c z^{2} = 0$ with $a, b, c$ coprime squarefree integers of mixed sign has a non-zero rational solution iff each of $- b c$ , $- c a$ , $- ab$ is a quadratic residue modulo the absolute value of the third coefficient. Legendre's conditions are exactly the local solvability conditions at each prime dividing the discriminant, although the language of local fields was a century in the future. Gauss's 1801 Disquisitiones Arithmeticae §V refined the binary case via genus theory and quadratic reciprocity; the genus class of a binary form is the collection of local equivalence classes at every prime, and Gauss's genus theorem is a local-global statement avant la lettre. Minkowski's 1890 paper Über die Bedingungen, unter welchen zwei quadratische Formen mit rationalen Coefficienten ineinander rational transformiert werden können (J. reine angew. Math. 106, 5-26) ^{[Minkowski 1890]} gave the first local-global classification of rational quadratic forms in $n$ variables, using the symbol $C_{p} (f)$ that is the modern Hasse-Witt invariant $ε_{p} (f)$ rewritten in pre-Hilbert-symbol notation. Minkowski's framework was algebraic and pre- $p$ -adic; the local condition at $p$ was a congruence condition rather than a $Q_{p}$ -statement, but the structure was already correct.

Hasse's 1923-24 papers Über die Äquivalenz quadratischer Formen (J. reine angew. Math. 152, 205-224) ^{[Hasse 1923]} and Darstellbarkeit von Zahlen durch quadratische Formen (J. reine angew. Math. 153, 12-43) ^{[Hasse 1924]} extended Minkowski's classification from equivalence to representation, introduced the modern $p$ -adic completion language following Hensel's 1908 Theorie der algebraischen Zahlen, and proved the representation local-global theorem now known as Hasse-Minkowski. Hasse's papers also contain a proof of Meyer's theorem in the modern $p$ -adic framework. The combined Minkowski-Hasse formulation became the textbook statement after Witt's 1937 Theorie der quadratischen Formen in beliebigen Körpern (J. reine angew. Math. 176, 31-44) introduced Witt cancellation and packaged the equivalence theory in the form Serre 1973 Ch. IV uses.

Selmer's 1951 The Diophantine equation $a x^{3} + b y^{3} + c z^{3} = 0$ (Acta Math. 85, 203-362) ^{[Selmer 1951]} produced the first known counterexample to the Hasse principle outside the quadratic-form setting, exhibiting the cubic curve $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ with rational points at every $Q_{v}$ but no non-zero rational point. The Selmer example crystallised the question of how to detect failures of the Hasse principle for higher-degree forms, and Manin's 1971 ICM Nice address introduced the Brauer-Manin obstruction as a cohomological refinement; Skorobogatov 1999 later showed that Brauer-Manin is itself sometimes insufficient and that higher étale-Brauer obstructions arise. The arc from Legendre 1785 to Manin 1971 traces the development of a single idea — that local conditions at every place control global solvability — from elementary congruences through $p$ -adic completions to Galois cohomology.

Bibliography Master

@article{Legendre1785,
  author  = {Legendre, Adrien-Marie},
  title   = {Recherches d'analyse ind{\'e}termin{\'e}e},
  journal = {Histoire de l'Acad{\'e}mie Royale des Sciences},
  year    = {1785},
  pages   = {465--559}
}

@book{Gauss1801,
  author    = {Gauss, Carl Friedrich},
  title     = {Disquisitiones Arithmeticae},
  publisher = {Gerhard Fleischer},
  address   = {Leipzig},
  year      = {1801}
}

@article{Minkowski1890,
  author  = {Minkowski, Hermann},
  title   = {{\"U}ber die Bedingungen, unter welchen zwei quadratische Formen mit rationalen Coefficienten ineinander rational transformiert werden k{\"o}nnen},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {106},
  year    = {1890},
  pages   = {5--26}
}

@book{Hensel1908,
  author    = {Hensel, Kurt},
  title     = {Theorie der algebraischen Zahlen},
  publisher = {Teubner},
  address   = {Leipzig},
  year      = {1908}
}

@article{Hasse1923,
  author  = {Hasse, Helmut},
  title   = {{\"U}ber die {\"A}quivalenz quadratischer Formen im K{\"o}rper der rationalen Zahlen},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {152},
  year    = {1923},
  pages   = {205--224}
}

@article{Hasse1924,
  author  = {Hasse, Helmut},
  title   = {Darstellbarkeit von Zahlen durch quadratische Formen im K{\"o}rper der rationalen Zahlen},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {153},
  year    = {1924},
  pages   = {12--43}
}

@article{Meyer1884,
  author  = {Meyer, Arnold},
  title   = {{\"U}ber die Aufl{\"o}sung der Gleichung $ax^2 + by^2 + cz^2 + du^2 + ev^2 = 0$ in ganzen Zahlen},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {98},
  year    = {1884},
  pages   = {177--230}
}

@article{Witt1937,
  author  = {Witt, Ernst},
  title   = {Theorie der quadratischen Formen in beliebigen K{\"o}rpern},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {176},
  year    = {1937},
  pages   = {31--44}
}

@article{Dirichlet1837,
  author  = {Dirichlet, Peter Gustav Lejeune},
  title   = {Beweis des Satzes, dass jede unbegrenzte arithmetische Progression, deren erstes Glied und Differenz ganze Zahlen ohne gemeinschaftlichen Factor sind, unendlich viele Primzahlen enth{\"a}lt},
  journal = {Abhandlungen der K{\"o}niglich Preussischen Akademie der Wissenschaften zu Berlin},
  year    = {1837},
  pages   = {45--81}
}

@book{Serre1973,
  author    = {Serre, Jean-Pierre},
  title     = {A Course in Arithmetic},
  publisher = {Springer-Verlag},
  series    = {Graduate Texts in Mathematics},
  volume    = {7},
  year      = {1973},
  note      = {English translation of \emph{Cours d'arithm{\'e}tique}, Presses Universitaires de France, 1970}
}

@book{Cassels1978,
  author    = {Cassels, J. W. S.},
  title     = {Rational Quadratic Forms},
  publisher = {Academic Press},
  series    = {London Mathematical Society Monographs},
  volume    = {13},
  year      = {1978}
}

@book{OMeara1963,
  author    = {O'Meara, O. Timothy},
  title     = {Introduction to Quadratic Forms},
  publisher = {Springer-Verlag},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {117},
  year      = {1963}
}

@book{BorevichShafarevich1966,
  author    = {Borevich, Z. I. and Shafarevich, I. R.},
  title     = {Number Theory},
  publisher = {Academic Press},
  year      = {1966},
  note      = {Translated by Newcomb Greenleaf from the 1964 Russian original}
}

@book{Lam2005,
  author    = {Lam, T. Y.},
  title     = {Introduction to Quadratic Forms over Fields},
  publisher = {American Mathematical Society},
  series    = {Graduate Studies in Mathematics},
  volume    = {67},
  year      = {2005}
}

@article{Selmer1951,
  author  = {Selmer, Ernst S.},
  title   = {The Diophantine equation $ax^3 + by^3 + cz^3 = 0$},
  journal = {Acta Mathematica},
  volume  = {85},
  year    = {1951},
  pages   = {203--362}
}

@inproceedings{Manin1971,
  author    = {Manin, Yu. I.},
  title     = {Le groupe de {B}rauer-{G}rothendieck en g{\'e}om{\'e}trie diophantienne},
  booktitle = {Actes du Congr{\`e}s International des Math{\'e}maticiens, Nice 1970},
  volume    = {1},
  publisher = {Gauthier-Villars},
  address   = {Paris},
  year      = {1971},
  pages     = {401--411}
}

@article{Skorobogatov1999,
  author  = {Skorobogatov, Alexei N.},
  title   = {Beyond the {M}anin obstruction},
  journal = {Inventiones Mathematicae},
  volume  = {135},
  year    = {1999},
  pages   = {399--424}
}

@article{Kneser1965,
  author  = {Kneser, Martin},
  title   = {Galois-{K}ohomologie halbeinfacher algebraischer {G}ruppen {\"u}ber $\mathfrak{p}$-adischen K{\"o}rpern},
  journal = {Mathematische Zeitschrift},
  volume  = {89},
  year    = {1965},
  pages   = {250--272}
}

@incollection{Harder1968,
  author    = {Harder, G{\"u}nter},
  title     = {Bericht {\"u}ber neuere {R}esultate der {G}aloiskohomologie halbeinfacher {G}ruppen},
  booktitle = {Jahresbericht der Deutschen Mathematiker-Vereinigung},
  volume    = {70},
  year      = {1968},
  pages     = {182--216}
}

Prerequisites

01.01.01
01.01.15
02.01.05
21.03.01

Tier anchors

beginner: Serre 1973 *A Course in Arithmetic* Ch. IV §3 informal opening; Conway 1997 *The Sensual (Quadratic) Form* (MAA Carus Monograph 26) Ch. 1-3 informal opening
intermediate: Serre 1973 *A Course in Arithmetic* (Springer GTM 7) Ch. IV §§1-3; Cassels 1978 *Rational Quadratic Forms* (Academic Press, LMS Monograph 13) Ch. 6; Borevich-Shafarevich 1966 *Number Theory* (Academic Press) Ch. 1 §7
master: Minkowski 1890 *J. reine angew. Math.* 106 (originator, rational forms); Hasse 1923 *J. reine angew. Math.* 152 + Hasse 1924 *J. reine angew. Math.* 153 (representation local-global, modern statement); Serre 1973 *A Course in Arithmetic* Ch. III-IV (the canonical short modern proof using the Hilbert symbol product formula); Cassels 1978 *Rational Quadratic Forms* Ch. 6 (full textbook proof); O'Meara 1963 *Introduction to Quadratic Forms* (Springer Grundlehren 117) Ch. VI (general global-field setting); Lam 2005 *Introduction to Quadratic Forms over Fields* (AMS GSM 67) Ch. VI; Meyer 1884 *J. reine angew. Math.* 98 (indefinite forms in five or more variables); Selmer 1951 *Acta Mathematica* 85 (cubic counterexample); Manin 1971 ICM Nice (Brauer-Manin obstruction)

References

Serre, J.-P. — A Course in Arithmetic · Springer Graduate Texts in Mathematics 7 (1973), translation of *Cours d'arithmétique* (PUF 1970). Ch. II (p-adic fields); Ch. III (Hilbert symbol and product formula); Ch. IV §§1-3 (quadratic forms over $\mathbb{Q}_p$ and $\mathbb{Q}$; Hasse-Minkowski theorem); Ch. IV §4 (Meyer's theorem for indefinite forms in $\geq 5$ variables).
Minkowski, H. — Über die Bedingungen, unter welchen zwei quadratische Formen mit rationalen Coefficienten ineinander rational transformiert werden können · *Journal für die reine und angewandte Mathematik* 106 (1890), 5-26. The originator paper for the rational-equivalence local-global theorem: two non-degenerate quadratic forms over $\mathbb{Q}$ are rationally equivalent iff they are equivalent over $\mathbb{R}$ and over every $p$-adic local invariant Minkowski introduced.
Hasse, H. — Über die Äquivalenz quadratischer Formen im Körper der rationalen Zahlen · *Journal für die reine und angewandte Mathematik* 152 (1923), 205-224. The first half of the Hasse extension: a quadratic form over $\mathbb{Q}$ is determined up to equivalence by its local invariants at every completion $\mathbb{Q}_v$ including $v = \infty$; the modern statement using $\mathbb{Q}_p$.
Hasse, H. — Darstellbarkeit von Zahlen durch quadratische Formen im Körper der rationalen Zahlen · *Journal für die reine und angewandte Mathematik* 153 (1924), 12-43. The representation theorem: a rational quadratic form represents a given rational number iff it represents that number over every completion. Includes Meyer's theorem on indefinite forms in $\geq 5$ variables.
Meyer, A. — Über die Auflösung der Gleichung $ax^2 + by^2 + cz^2 + du^2 + ev^2 = 0$ in ganzen Zahlen · *Journal für die reine und angewandte Mathematik* 98 (1884), 177-230. Meyer's theorem: every indefinite quadratic form in $\geq 5$ variables with rational coefficients represents zero by a non-zero vector over $\mathbb{Q}$.
Selmer, E. S. — The Diophantine equation $ax^3 + by^3 + cz^3 = 0$ · *Acta Mathematica* 85 (1951), 203-362. Contains the canonical counterexample $3 x^3 + 4 y^3 + 5 z^3 = 0$: a cubic curve with rational points over every $\mathbb{Q}_v$ but no non-zero rational point, demonstrating that the Hasse principle fails for higher-degree forms.
Manin, Yu. I. — Le groupe de Brauer-Grothendieck en géométrie diophantienne · *Actes du Congrès International des Mathématiciens* (Nice, 1970), Vol. 1, Gauthier-Villars (Paris, 1971), pp. 401-411. Introduces the Brauer-Manin obstruction: a cohomological refinement of the adelic points that detects the Selmer counterexample and many other failures of the Hasse principle.
Cassels, J. W. S. — Rational Quadratic Forms · Academic Press, London Mathematical Society Monograph 13 (1978). Ch. 6 contains the full textbook-depth proof of the Hasse-Minkowski theorem, with the genus theory background and a careful treatment of the $n = 4$ counting argument.
O'Meara, O. T. — Introduction to Quadratic Forms · Springer Grundlehren der mathematischen Wissenschaften 117 (1963), reprinted 2000. Ch. VI develops the local-global theory of quadratic forms over arbitrary global fields, generalising the Hasse-Minkowski theorem beyond $\mathbb{Q}$.
Borevich, Z. I. and Shafarevich, I. R. — Number Theory · Academic Press (1966), trans. N. Greenleaf from the 1964 Russian original. Ch. 1 §7 develops Hasse-Minkowski in the Soviet-school style with the symbol $(\alpha, \beta)_p$ computations explicit.
Lam, T. Y. — Introduction to Quadratic Forms over Fields · American Mathematical Society Graduate Studies in Mathematics 67 (2005). Ch. VI gives a modern algebraic treatment of the Hasse principle with the Witt-ring framing.
Dirichlet, P. G. L. — Beweis des Satzes, dass jede unbegrenzte arithmetische Progression, deren erstes Glied und Differenz ganze Zahlen ohne gemeinschaftlichen Factor sind, unendlich viele Primzahlen enthält · *Abhandlungen der Königlich Preussischen Akademie der Wissenschaften zu Berlin* (1837), 45-81. Originator paper for Dirichlet's theorem on primes in arithmetic progressions, the analytic input the Hasse-Minkowski proof at $n = 3$ relies on.
Legendre, A. M. — Recherches d'analyse indéterminée · *Histoire de l'Académie Royale des Sciences* (Paris, 1785), 465-559. The three-variable case $ax^2 + by^2 + cz^2 = 0$ — the historical seed of Hasse-Minkowski, with Legendre's congruence conditions reading as a local-global statement avant la lettre.

Estimated time

beginner: 18m
intermediate: 45m
master: 90m